Waterfall Enrichment for People

Outcome: work emails on your outbound list with a measured hit rate, misses routed to a fallback channel, and the column running only on qualified rows.

Surface: App and MCP server
Level: Intermediate
Uses: find_work_email
Credits: ~1 per hit
Prerequisite: A qualified person list from lessons 01–04

The gate comes first

This is the most expensive column in an outbound build. Before it runs, three things must be true:

The company passed your ICP filter.
The person matches the buying role, not just any role at the company.
The row is not suppressed — not a customer, not an open opportunity, not contacted recently.

Each of those is a free filter. Together they typically remove 60–80% of rows, and every removed row is a credit you keep.

Input hygiene lifts hit rate more than anything else

Split names properly

First and last as separate fields, honorifics and suffixes stripped. Combined name fields cost hits.

Use the operating domain

Not the parent, not the group site. If the person works at the UK subsidiary, the parent domain will not find them.

Normalize the domain

Lowercase, no protocol, no www., no path.

Pass the LinkedIn URL where you have it

It is an exact identifier and outperforms name-plus-domain.

These four move a hit rate by 10–20 points. No provider change comes close.

Expected hit rates

Segment	Hit rate
Corporate B2B, 200+ employees	75–90%
Mid-market, 50–200	65–80%
Small business, under 50	40–60%
Local business, sole traders	20–40%
Non-English markets	Varies widely — measure yours

If your number is well below the band for your segment, check inputs before assuming coverage. On a first run, most misses are input errors.

Reference: Find work email, find_work_email.

Route the misses

A miss is not a dead row. Three fallbacks, in order:

Fallback	When	Cost
Site scraping for a contact address	The company has a website and is small enough that a shared inbox is read	Cheap
Phone	Local, SMB, or a high-value account	Expensive — gate it
LinkedIn	The person is reachable there and the volume is small	Manual

And a fourth option that is often correct: accept the miss. A row with no contact path is a row that does not enter the campaign, and forcing coverage on it is usually more expensive than the row is worth.

Record contact_source on every row — waterfall, scraped, phone. It is the only way to compare reply rates by source later, and that comparison is what tells you whether the fallbacks are worth their cost.

Do this now

Confirm the three gate conditions

Company qualified, role matched, not suppressed.

Fix the inputs

Names split, domains normalized and operating-entity, LinkedIn URLs attached.

Run on ten people you can verify

Colleagues or customers whose addresses you know.

Fix any input errors the test reveals

Run across the gated list

Compute hit rate against rows attempted

Not rows in the table.

Route the misses

Scraping for small companies, phone for high-value, accept the rest.

Record contact_source per row

Check your work

The column ran behind all three gates
Names split, domains are operating entities
You tested against known addresses first
Hit rate computed against rows attempted
Misses are routed or explicitly accepted
contact_source is populated

Where this breaks

Running the waterfall before the ICP filter is the most expensive mistake in outbound, and it is common because contacts feel like the goal. On an ungated list most of the spend lands on companies you would have dropped for free — and the resulting list is both more expensive and worse, because it dilutes the good rows with rows nobody will work.

Further automation

Log hit rate per run and per segment. A falling hit rate is usually the first visible sign that your sourcing drifted into a segment with thinner coverage — which is a targeting problem you want to catch before it shows up as fewer meetings.

Next lesson

06 — Verify before you send, the step that decides which of these addresses are real.

Reference for this lesson: Find work email, find_work_email, Waterfall Enrichment, Credits.